2 * Copyright (c) 1989, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)nfs_vfsops.c 8.12 (Berkeley) 5/20/95
37 * $FreeBSD: src/sys/nfs/nfs_vfsops.c,v 1.91.2.7 2003/01/27 20:04:08 dillon Exp $
38 * $DragonFly: src/sys/vfs/nfs/nfs_vfsops.c,v 1.37 2006/03/27 17:01:18 dillon Exp $
41 #include "opt_bootp.h"
42 #include "opt_nfsroot.h"
44 #include <sys/param.h>
45 #include <sys/sockio.h>
47 #include <sys/vnode.h>
48 #include <sys/kernel.h>
49 #include <sys/sysctl.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/systm.h>
58 #include <vm/vm_extern.h>
59 #include <vm/vm_zone.h>
62 #include <net/route.h>
63 #include <netinet/in.h>
65 #include <sys/thread2.h>
73 #include "nfsm_subs.h"
74 #include "nfsdiskless.h"
75 #include "nfsmountrpc.h"
77 extern int nfs_mountroot(struct mount *mp);
78 extern void bootpc_init(void);
81 extern struct vnodeopv_entry_desc nfsv2_vnodeop_entries[];
82 extern struct vnodeopv_entry_desc nfsv2_fifoop_entries[];
83 extern struct vnodeopv_entry_desc nfsv2_specop_entries[];
85 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
86 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
87 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
88 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
89 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
90 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
91 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
93 vm_zone_t nfsmount_zone;
95 struct nfsstats nfsstats;
96 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
97 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
98 &nfsstats, nfsstats, "");
99 static int nfs_ip_paranoia = 1;
100 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
101 &nfs_ip_paranoia, 0, "");
104 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
108 * Tunable to determine the Read/Write unit size. Maximum value
109 * is NFS_MAXDATA. We also default to NFS_MAXDATA.
111 static int nfs_io_size = NFS_MAXDATA;
112 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
113 &nfs_io_size, 0, "NFS optimal I/O unit size");
115 static void nfs_decode_args (struct nfsmount *nmp,
116 struct nfs_args *argp);
117 static int mountnfs (struct nfs_args *,struct mount *,
118 struct sockaddr *,char *,char *,struct vnode **);
119 static int nfs_mount ( struct mount *mp, char *path, caddr_t data,
121 static int nfs_unmount ( struct mount *mp, int mntflags,
123 static int nfs_root ( struct mount *mp, struct vnode **vpp);
124 static int nfs_statfs ( struct mount *mp, struct statfs *sbp,
126 static int nfs_sync ( struct mount *mp, int waitfor,
130 * nfs vfs operations.
132 static struct vfsops nfs_vfsops = {
133 .vfs_mount = nfs_mount,
134 .vfs_unmount = nfs_unmount,
135 .vfs_root = nfs_root,
136 .vfs_statfs = nfs_statfs,
137 .vfs_sync = nfs_sync,
138 .vfs_init = nfs_init,
139 .vfs_uninit = nfs_uninit
141 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
144 * This structure must be filled in by a primary bootstrap or bootstrap
145 * server for a diskless/dataless machine. It is initialized below just
146 * to ensure that it is allocated to initialized data (.data not .bss).
148 struct nfs_diskless nfs_diskless = { { { 0 } } };
149 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
150 int nfs_diskless_valid = 0;
152 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
153 &nfs_diskless_valid, 0, "");
155 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
156 nfsv3_diskless.root_hostnam, 0, "");
158 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
159 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
160 "%Ssockaddr_in", "");
162 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
163 nfsv3_diskless.swap_hostnam, 0, "");
165 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
166 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
170 void nfsargs_ntoh (struct nfs_args *);
171 static int nfs_mountdiskless (char *, char *, int,
172 struct sockaddr_in *, struct nfs_args *,
173 struct thread *, struct vnode **,
175 static void nfs_convert_diskless (void);
176 static void nfs_convert_oargs (struct nfs_args *args,
177 struct onfs_args *oargs);
180 * Calculate the buffer I/O block size to use. The maximum V2 block size
181 * is typically 8K, the maximum datagram size is typically 16K, and the
182 * maximum V3 block size is typically 32K. The buffer cache tends to work
183 * best with 16K blocks but we allow 32K for TCP connections.
185 * We force the block size to be at least a page for buffer cache efficiency.
189 nfs_iosize(int v3, int sotype)
195 if (sotype == SOCK_STREAM)
198 iomax = NFS_MAXDGRAMDATA;
200 iomax = NFS_V2MAXDATA;
202 if ((iosize = nfs_io_size) > iomax)
204 if (iosize < PAGE_SIZE)
208 * This is an aweful hack but until the buffer cache is rewritten
209 * we need it. The problem is that when you combine write() with
210 * mmap() the vm_page->valid bits can become weird looking
211 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers
212 * at the file EOF. To solve the problem the BIO system needs to
213 * be guarenteed that the NFS iosize for regular files will be a
214 * multiple of PAGE_SIZE so it can invalidate the whole page
215 * rather then just the piece of it owned by the buffer when
216 * NFS does vinvalbuf() calls.
218 if (iosize & PAGE_MASK)
219 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
224 nfs_convert_oargs(args, oargs)
225 struct nfs_args *args;
226 struct onfs_args *oargs;
228 args->version = NFS_ARGSVERSION;
229 args->addr = oargs->addr;
230 args->addrlen = oargs->addrlen;
231 args->sotype = oargs->sotype;
232 args->proto = oargs->proto;
233 args->fh = oargs->fh;
234 args->fhsize = oargs->fhsize;
235 args->flags = oargs->flags;
236 args->wsize = oargs->wsize;
237 args->rsize = oargs->rsize;
238 args->readdirsize = oargs->readdirsize;
239 args->timeo = oargs->timeo;
240 args->retrans = oargs->retrans;
241 args->maxgrouplist = oargs->maxgrouplist;
242 args->readahead = oargs->readahead;
243 args->deadthresh = oargs->deadthresh;
244 args->hostname = oargs->hostname;
248 nfs_convert_diskless()
252 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
253 sizeof(struct ifaliasreq));
254 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
255 sizeof(struct sockaddr_in));
256 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
258 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
259 nfsv3_diskless.swap_fhsize = NFSX_V2FH;
260 for (i = NFSX_V2FH - 1; i >= 0; --i) {
261 if (nfs_diskless.swap_fh[i])
265 nfsv3_diskless.swap_fhsize = 0;
267 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
268 sizeof(struct sockaddr_in));
269 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
270 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
271 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
272 sizeof(struct ucred));
273 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
275 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
276 nfsv3_diskless.root_fhsize = NFSX_V2FH;
277 for (i = NFSX_V2FH - 1; i >= 0; --i) {
278 if (nfs_diskless.root_fh[i])
282 nfsv3_diskless.root_fhsize = 0;
284 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
285 sizeof(struct sockaddr_in));
286 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
287 nfsv3_diskless.root_time = nfs_diskless.root_time;
288 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
290 nfs_diskless_valid = 3;
297 nfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
300 struct nfs_statfs *sfp;
304 caddr_t bpos, dpos, cp2;
305 struct nfsmount *nmp = VFSTONFS(mp);
306 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
307 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
313 sfp = (struct nfs_statfs *)0;
315 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
320 cred->cr_ngroups = 1;
321 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
322 (void)nfs_fsinfo(nmp, vp, td);
323 nfsstats.rpccnt[NFSPROC_FSSTAT]++;
324 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
326 nfsm_request(vp, NFSPROC_FSSTAT, td, cred);
328 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
334 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
335 sbp->f_flags = nmp->nm_flag;
336 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype);
339 sbp->f_bsize = NFS_FABLKSIZE;
340 tquad = fxdr_hyper(&sfp->sf_tbytes);
341 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
342 tquad = fxdr_hyper(&sfp->sf_fbytes);
343 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
344 tquad = fxdr_hyper(&sfp->sf_abytes);
345 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
346 sbp->f_files = (fxdr_unsigned(int32_t,
347 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
348 sbp->f_ffree = (fxdr_unsigned(int32_t,
349 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
351 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
352 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
353 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
354 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
358 if (sbp != &mp->mnt_stat) {
359 sbp->f_type = mp->mnt_vfc->vfc_typenum;
360 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
370 * nfs version 3 fsinfo rpc call
373 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
375 struct nfsv3_fsinfo *fsp;
378 u_int32_t *tl, pref, max;
379 caddr_t bpos, dpos, cp2;
380 int error = 0, retattr;
381 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
384 nfsstats.rpccnt[NFSPROC_FSINFO]++;
385 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
387 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ));
388 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
390 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
391 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
392 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
393 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
394 ~(NFS_FABLKSIZE - 1);
395 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
396 if (max < nmp->nm_wsize && max > 0) {
397 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
398 if (nmp->nm_wsize == 0)
401 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
402 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
403 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
404 ~(NFS_FABLKSIZE - 1);
405 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
406 if (max < nmp->nm_rsize && max > 0) {
407 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
408 if (nmp->nm_rsize == 0)
411 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
412 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
413 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
414 ~(NFS_DIRBLKSIZ - 1);
415 if (max < nmp->nm_readdirsize && max > 0) {
416 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
417 if (nmp->nm_readdirsize == 0)
418 nmp->nm_readdirsize = max;
420 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
421 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
422 nmp->nm_maxfilesize = maxfsize;
423 nmp->nm_state |= NFSSTA_GOTFSINFO;
431 * Mount a remote root fs via. nfs. This depends on the info in the
432 * nfs_diskless structure that has been filled in properly by some primary
434 * It goes something like this:
435 * - do enough of "ifconfig" by calling ifioctl() so that the system
436 * can talk to the server
437 * - If nfs_diskless.mygateway is filled in, use that address as
439 * - build the rootfs mount point and call mountnfs() to do the rest.
445 struct mount *swap_mp;
446 struct nfsv3_diskless *nd = &nfsv3_diskless;
449 struct thread *td = curthread; /* XXX */
454 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
455 bootpc_init(); /* use bootp to get nfs_diskless filled in */
459 * XXX time must be non-zero when we init the interface or else
460 * the arp code will wedge...
462 while (mycpu->gd_time_seconds == 0)
463 tsleep(mycpu, 0, "arpkludge", 10);
466 * The boot code may have passed us a diskless structure.
468 if (nfs_diskless_valid == 1)
469 nfs_convert_diskless();
471 #define SINP(sockaddr) ((struct sockaddr_in *)(sockaddr))
472 printf("nfs_mountroot: interface %s ip %s",
474 inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
476 inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
478 inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
482 * XXX splnet, so networks will receive...
487 * BOOTP does not necessarily have to be compiled into the kernel
488 * for an NFS root to work. If we inherited the network
489 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
490 * out our interface for us and all we need to do is ifconfig the
491 * interface. We only do this if the interface has not already been
492 * ifconfig'd by e.g. BOOTP.
494 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
496 panic("nfs_mountroot: socreate(%04x): %d",
497 nd->myif.ifra_addr.sa_family, error);
500 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td);
502 panic("nfs_mountroot: SIOCAIFADDR: %d", error);
507 * If the gateway field is filled in, set it as the default route.
509 if (nd->mygateway.sin_len != 0) {
510 struct sockaddr_in mask, sin;
512 bzero((caddr_t)&mask, sizeof(mask));
514 sin.sin_family = AF_INET;
515 sin.sin_len = sizeof(sin);
516 printf("nfs_mountroot: gateway %s\n",
517 inet_ntoa(nd->mygateway.sin_addr));
518 error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
519 (struct sockaddr *)&nd->mygateway,
520 (struct sockaddr *)&mask,
521 RTF_UP | RTF_GATEWAY);
523 printf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
527 * Create the rootfs mount point.
529 nd->root_args.fh = nd->root_fh;
530 nd->root_args.fhsize = nd->root_fhsize;
531 l = ntohl(nd->root_saddr.sin_addr.s_addr);
532 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
533 (l >> 24) & 0xff, (l >> 16) & 0xff,
534 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam);
535 printf("NFS ROOT: %s\n",buf);
536 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
537 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
539 mp->mnt_vfc->vfc_refcount--;
540 free(swap_mp, M_MOUNT);
547 if (nd->swap_nblks) {
549 /* Convert to DEV_BSIZE instead of Kilobyte */
553 * Create a fake mount point just for the swap vnode so that the
554 * swap file can be on a different server from the rootfs.
556 nd->swap_args.fh = nd->swap_fh;
557 nd->swap_args.fhsize = nd->swap_fhsize;
558 l = ntohl(nd->swap_saddr.sin_addr.s_addr);
559 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
560 (l >> 24) & 0xff, (l >> 16) & 0xff,
561 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam);
562 printf("NFS SWAP: %s\n",buf);
563 if ((error = nfs_mountdiskless(buf, "/swap", 0,
564 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) {
568 vfs_unbusy(swap_mp, td);
570 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
571 nd->swap_nblks * DEV_BSIZE ;
574 * Since the swap file is not the root dir of a file system,
575 * hack it to a regular file.
580 swaponvp(td, vp, nd->swap_nblks);
583 mp->mnt_flag |= MNT_ROOTFS;
584 mp->mnt_vnodecovered = NULLVP;
588 * This is not really an nfs issue, but it is much easier to
589 * set hostname here and then let the "/etc/rc.xxx" files
590 * mount the right /var based upon its preset value.
592 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
593 hostname[MAXHOSTNAMELEN - 1] = '\0';
594 for (i = 0; i < MAXHOSTNAMELEN; i++)
595 if (hostname[i] == '\0')
597 inittodr(ntohl(nd->root_time));
603 * Internal version of mount system call for diskless setup.
606 nfs_mountdiskless(char *path, char *which, int mountflag,
607 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
608 struct vnode **vpp, struct mount **mpp)
611 struct sockaddr *nam;
618 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
619 printf("nfs_mountroot: NFS not configured");
624 mp->mnt_kern_flag = 0;
625 mp->mnt_flag = mountflag;
626 nam = dup_sockaddr((struct sockaddr *)sin);
628 #if defined(BOOTP) || defined(NFS_ROOT)
629 if (args->fhsize == 0) {
630 printf("NFS_ROOT: No FH passed from loader, attempting mount rpc...");
632 error = md_mount(sin, which, args->fh, &args->fhsize, args, td);
637 printf("success!\n");
641 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
642 #if defined(BOOTP) || defined(NFS_ROOT)
645 printf("nfs_mountroot: mount %s on %s: %d", path, which, error);
646 mp->mnt_vfc->vfc_refcount--;
658 nfs_decode_args(nmp, argp)
659 struct nfsmount *nmp;
660 struct nfs_args *argp;
667 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
668 * no sense in that context.
670 if (argp->sotype == SOCK_STREAM)
671 nmp->nm_flag &= ~NFSMNT_NOCONN;
673 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
674 if ((argp->flags & NFSMNT_NFSV3) == 0)
675 nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
677 /* Re-bind if rsrvd port requested and wasn't on one */
678 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
679 && (argp->flags & NFSMNT_RESVPORT);
680 /* Also re-bind if we're switching to/from a connected UDP socket */
681 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
682 (argp->flags & NFSMNT_NOCONN));
684 /* Update flags atomically. Don't change the lock bits. */
685 nmp->nm_flag = argp->flags | nmp->nm_flag;
688 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
689 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
690 if (nmp->nm_timeo < NFS_MINTIMEO)
691 nmp->nm_timeo = NFS_MINTIMEO;
692 else if (nmp->nm_timeo > NFS_MAXTIMEO)
693 nmp->nm_timeo = NFS_MAXTIMEO;
696 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
697 nmp->nm_retry = argp->retrans;
698 if (nmp->nm_retry > NFS_MAXREXMIT)
699 nmp->nm_retry = NFS_MAXREXMIT;
702 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
704 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
705 nmp->nm_wsize = argp->wsize;
706 /* Round down to multiple of blocksize */
707 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
708 if (nmp->nm_wsize <= 0)
709 nmp->nm_wsize = NFS_FABLKSIZE;
711 if (nmp->nm_wsize > maxio)
712 nmp->nm_wsize = maxio;
713 if (nmp->nm_wsize > MAXBSIZE)
714 nmp->nm_wsize = MAXBSIZE;
716 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
717 nmp->nm_rsize = argp->rsize;
718 /* Round down to multiple of blocksize */
719 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
720 if (nmp->nm_rsize <= 0)
721 nmp->nm_rsize = NFS_FABLKSIZE;
723 if (nmp->nm_rsize > maxio)
724 nmp->nm_rsize = maxio;
725 if (nmp->nm_rsize > MAXBSIZE)
726 nmp->nm_rsize = MAXBSIZE;
728 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
729 nmp->nm_readdirsize = argp->readdirsize;
731 if (nmp->nm_readdirsize > maxio)
732 nmp->nm_readdirsize = maxio;
733 if (nmp->nm_readdirsize > nmp->nm_rsize)
734 nmp->nm_readdirsize = nmp->nm_rsize;
736 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
737 nmp->nm_acregmin = argp->acregmin;
739 nmp->nm_acregmin = NFS_MINATTRTIMO;
740 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
741 nmp->nm_acregmax = argp->acregmax;
743 nmp->nm_acregmax = NFS_MAXATTRTIMO;
744 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
745 nmp->nm_acdirmin = argp->acdirmin;
747 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
748 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
749 nmp->nm_acdirmax = argp->acdirmax;
751 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
752 if (nmp->nm_acdirmin > nmp->nm_acdirmax)
753 nmp->nm_acdirmin = nmp->nm_acdirmax;
754 if (nmp->nm_acregmin > nmp->nm_acregmax)
755 nmp->nm_acregmin = nmp->nm_acregmax;
757 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
758 if (argp->maxgrouplist <= NFS_MAXGRPS)
759 nmp->nm_numgrps = argp->maxgrouplist;
761 nmp->nm_numgrps = NFS_MAXGRPS;
763 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
764 if (argp->readahead <= NFS_MAXRAHEAD)
765 nmp->nm_readahead = argp->readahead;
767 nmp->nm_readahead = NFS_MAXRAHEAD;
769 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
770 if (argp->deadthresh <= NFS_NEVERDEAD)
771 nmp->nm_deadthresh = argp->deadthresh;
773 nmp->nm_deadthresh = NFS_NEVERDEAD;
776 adjsock |= ((nmp->nm_sotype != argp->sotype) ||
777 (nmp->nm_soproto != argp->proto));
778 nmp->nm_sotype = argp->sotype;
779 nmp->nm_soproto = argp->proto;
781 if (nmp->nm_so && adjsock) {
782 nfs_safedisconnect(nmp);
783 if (nmp->nm_sotype == SOCK_DGRAM)
784 while (nfs_connect(nmp, (struct nfsreq *)0)) {
785 printf("nfs_args: retrying connect\n");
786 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
795 * It seems a bit dumb to copyinstr() the host and path here and then
796 * bcopy() them in mountnfs(), but I wanted to detect errors before
797 * doing the sockargs() call because sockargs() allocates an mbuf and
798 * an error after that means that I have to release the mbuf.
802 nfs_mount(struct mount *mp, char *path, caddr_t data, struct thread *td)
805 struct nfs_args args;
806 struct sockaddr *nam;
808 char pth[MNAMELEN], hst[MNAMELEN];
810 u_char nfh[NFSX_V3FHMAX];
816 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
819 if (args.version != NFS_ARGSVERSION) {
820 #ifdef COMPAT_PRELITE2
822 * If the argument version is unknown, then assume the
823 * caller is a pre-lite2 4.4BSD client and convert its
826 struct onfs_args oargs;
827 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
830 nfs_convert_oargs(&args,&oargs);
831 #else /* !COMPAT_PRELITE2 */
832 return (EPROGMISMATCH);
833 #endif /* COMPAT_PRELITE2 */
835 if (mp->mnt_flag & MNT_UPDATE) {
836 struct nfsmount *nmp = VFSTONFS(mp);
841 * When doing an update, we can't change from or to
842 * v3, or change cookie translation
844 args.flags = (args.flags &
845 ~(NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/)) |
847 (NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/));
848 nfs_decode_args(nmp, &args);
853 * Make the nfs_ip_paranoia sysctl serve as the default connection
854 * or no-connection mode for those protocols that support
855 * no-connection mode (the flag will be cleared later for protocols
856 * that do not support no-connection mode). This will allow a client
857 * to receive replies from a different IP then the request was
858 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid),
861 if (nfs_ip_paranoia == 0)
862 args.flags |= NFSMNT_NOCONN;
863 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
865 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
868 error = copyinstr(path, pth, MNAMELEN-1, &len);
871 bzero(&pth[len], MNAMELEN - len);
872 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
875 bzero(&hst[len], MNAMELEN - len);
876 /* sockargs() call must be after above copyin() calls */
877 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
881 error = mountnfs(&args, mp, nam, pth, hst, &vp);
886 * Common code for mount and mountroot
889 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
890 char *pth, char *hst, struct vnode **vpp)
892 struct nfsmount *nmp;
896 if (mp->mnt_flag & MNT_UPDATE) {
898 /* update paths, file handles, etc, here XXX */
902 nmp = zalloc(nfsmount_zone);
903 bzero((caddr_t)nmp, sizeof (struct nfsmount));
904 TAILQ_INIT(&nmp->nm_uidlruhead);
905 TAILQ_INIT(&nmp->nm_bioq);
906 mp->mnt_data = (qaddr_t)nmp;
912 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
913 * high, depending on whether we end up with negative offsets in
914 * the client or server somewhere. 2GB-1 may be safer.
916 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum
917 * that we can handle until we find out otherwise.
918 * XXX Our "safe" limit on the client is what we can store in our
919 * buffer cache using signed(!) block numbers.
921 if ((argp->flags & NFSMNT_NFSV3) == 0)
922 nmp->nm_maxfilesize = 0xffffffffLL;
924 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
926 nmp->nm_timeo = NFS_TIMEO;
927 nmp->nm_retry = NFS_RETRANS;
928 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
929 nmp->nm_rsize = nmp->nm_wsize;
930 nmp->nm_readdirsize = NFS_READDIRSIZE;
931 nmp->nm_numgrps = NFS_MAXGRPS;
932 nmp->nm_readahead = NFS_DEFRAHEAD;
933 nmp->nm_deadthresh = NFS_DEADTHRESH;
934 nmp->nm_fhsize = argp->fhsize;
935 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
936 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
938 /* Set up the sockets and per-host congestion */
939 nmp->nm_sotype = argp->sotype;
940 nmp->nm_soproto = argp->proto;
941 nmp->nm_cred = crhold(proc0.p_ucred);
943 nfs_decode_args(nmp, argp);
946 * For Connection based sockets (TCP,...) defer the connect until
947 * the first request, in case the server is not responding.
949 if (nmp->nm_sotype == SOCK_DGRAM &&
950 (error = nfs_connect(nmp, (struct nfsreq *)0)))
954 * This is silly, but it has to be set so that vinifod() works.
955 * We do not want to do an nfs_statfs() here since we can get
956 * stuck on a dead server and we are holding a lock on the mount
959 mp->mnt_stat.f_iosize =
960 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
963 * Install vop_ops for our vnops
965 vfs_add_vnodeops(mp, &mp->mnt_vn_norm_ops,
966 nfsv2_vnodeop_entries, 0);
967 vfs_add_vnodeops(mp, &mp->mnt_vn_spec_ops,
968 nfsv2_specop_entries, 0);
969 vfs_add_vnodeops(mp, &mp->mnt_vn_fifo_ops,
970 nfsv2_fifoop_entries, 0);
973 * A reference count is needed on the nfsnode representing the
974 * remote root. If this object is not persistent, then backward
975 * traversals of the mount point (i.e. "..") will not work if
976 * the nfsnode gets flushed out of the cache. Ufs does not have
977 * this problem, because one can identify root inodes by their
978 * number == ROOTINO (2).
980 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
986 * Retrieval of mountpoint attributes is delayed until nfs_rot
987 * or nfs_statfs are first called. This will happen either when
988 * we first traverse the mount point or if somebody does a df(1).
990 * NFSSTA_GOTFSINFO is used to flag if we have successfully
991 * retrieved mountpoint attributes. In the case of NFSv3 we
992 * also flag static fsinfo.
995 (*vpp)->v_type = VNON;
998 * Lose the lock but keep the ref.
1000 VOP_UNLOCK(*vpp, 0, curthread);
1004 nfs_disconnect(nmp);
1005 nfs_free_mount(nmp);
1006 FREE(nam, M_SONAME);
1011 * unmount system call
1014 nfs_unmount(struct mount *mp, int mntflags, struct thread *td)
1016 struct nfsmount *nmp;
1017 int error, flags = 0;
1019 if (mntflags & MNT_FORCE)
1020 flags |= FORCECLOSE;
1023 * Goes something like this..
1024 * - Call vflush() to clear out vnodes for this file system
1025 * - Close the socket
1026 * - Free up the data structures
1028 /* In the forced case, cancel any outstanding requests. */
1029 if (flags & FORCECLOSE) {
1030 error = nfs_nmcancelreqs(nmp);
1035 * Must handshake with nfs_clientd() if it is active. XXX
1037 nmp->nm_state |= NFSSTA_DISMINPROG;
1039 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1040 error = vflush(mp, 1, flags);
1042 nmp->nm_state &= ~NFSSTA_DISMINPROG;
1047 * We are now committed to the unmount.
1048 * For NQNFS, let the server daemon free the nfsmount structure.
1050 if (nmp->nm_flag & NFSMNT_KERB)
1051 nmp->nm_state |= NFSSTA_DISMNT;
1053 nfs_disconnect(nmp);
1054 FREE(nmp->nm_nam, M_SONAME);
1056 if ((nmp->nm_flag & NFSMNT_KERB) == 0)
1057 nfs_free_mount(nmp);
1062 nfs_free_mount(struct nfsmount *nmp)
1065 crfree(nmp->nm_cred);
1066 nmp->nm_cred = NULL;
1068 zfree(nfsmount_zone, nmp);
1072 * Return root of a filesystem
1080 struct nfsmount *nmp;
1086 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1092 * Get transfer parameters and root vnode attributes
1094 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1095 if (nmp->nm_flag & NFSMNT_NFSV3) {
1096 nfs_fsinfo(nmp, vp, curthread);
1097 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1099 if ((error = VOP_GETATTR(vp, &attrs, curthread)) == 0)
1100 nmp->nm_state |= NFSSTA_GOTFSINFO;
1104 if (vp->v_type == VNON)
1120 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1121 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1124 * Flush out the buffer cache
1128 nfs_sync(struct mount *mp, int waitfor, struct thread *td)
1130 struct scaninfo scaninfo;
1133 scaninfo.rescan = 0;
1135 scaninfo.waitfor = waitfor;
1136 scaninfo.allerror = 0;
1139 * Force stale buffer cache information to be flushed.
1142 while (error == 0 && scaninfo.rescan) {
1143 scaninfo.rescan = 0;
1144 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1145 nfs_sync_scan2, &scaninfo);
1152 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1154 struct scaninfo *info = data;
1156 if (VOP_ISLOCKED(vp, NULL) || RB_EMPTY(&vp->v_rbdirty_tree))
1158 if (info->waitfor == MNT_LAZY)
1165 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1167 struct scaninfo *info = data;
1170 error = VOP_FSYNC(vp, info->waitfor, info->td);
1172 info->allerror = error;